Qwen2.5-VL-7B-Instruct_DPE_v2

Paper | GitHub

Qwen2.5-VL-7B-Instruct_DPE_v2 is the second-iteration model evolved through the DPE framework, building upon the foundations of v1.

🌟 Model Overview

DPE uses a multi-agent system to diagnose failure patterns and generate high-quality targeted data. This version represents the second evolutionary cycle.

📊 Evaluation Results

Category	Benchmark	Base Model	DPE_v2 (Ours)	Improvement
STEM	MMMU	53.11	55.33	+2.22
	RealWorldQA	68.63	69.54	+0.91
Visual Math	MathVista	65.50	68.20	+2.70
Specialized	HallusionBench	64.98	69.19	+4.21
Overall	Average	57.29	58.68	+1.39

📑 Citation

@misc{jia2026blindspotsgainsdiagnosticdriven,
      title={From Blind Spots to Gains: Diagnostic-Driven Iterative Training for Large Multimodal Models}, 
      author={Hongrui Jia and Chaoya Jiang and Shikun Zhang and Wei Ye},
      year={2026},
      eprint={2602.22859},
      archivePrefix={arXiv},
      primaryClass={cs.CV},
      url={https://arxiv.org/abs/2602.22859}, 
}

📜 License

This model follows the Qwen Research License.